Skip to content
Finalist

Astra(l) Projection

Astra(l) Projection turns real-world hardware into interactive 3D explanations, combining on-demand generation, lazy-loaded component detail, voice interaction, and native spatial rendering.

Code Video Cerebral Valley

Core trick

Reconstructs an object as selectable 3D component models from camera images and technical references, lazy-loading each part's detail.

Tags

visioneducation

Description

Astra(l) Projection turns real-world hardware into interactive 3D explanations, combining on-demand generation, lazy-loaded component detail, voice interaction, and native spatial rendering. In VR, users can reconstruct objects from camera images and technical references, pull apart assemblies, and explore their internals through guided walkthroughs. In AR, users can create and edit spatial models, visualize explanations through animated flows and labels, and generate illustrations.
abharw/astra-spatial★ 00 forksSwiftMITCreated Sep 8, 2026 · last push Oct 19, 2026

README excerpt

# Astra Spatial

Point, ask, and explore how things work in AR and VR. Built for the Cerebral Valley / OpenAI hackathon.

Built by [Arav Bhardwaj](https://github.com/abharw) and Akeil ([@akeildev](https://github.com/akeildev)). Arav built the iOS/iPadOS AR app (`app/AR`), the SpatialCore/SpatialApple Swift framework (`framework/`), and the Realtime backend (`backend/`). Akeil built the Quest 3 VR app (`app/VR`) and the rack asset pipeline.

## Features

- **Reconstruct real objects · VR.** Point at something and say “reconstruct that.” Astra identifies it from the camera image, searches technical references, and generates an approximate model with individually selectable components. The captured camera pose and surface depth help place it against the real-world surface.

- **Reveal detail on demand · AR + VR.** Start with an 18-server rack exterior and reveal its internal teaching groups as you explore. In VR, “show the processors in server 3” loads just that group. The model knows which parts are available before their geometry loads; bundled resources are decoded on demand, shared between instances, and released when unused.

- **Go deeper inside a component · VR.** Ask for more detail inside a generated part to create further selectable subcomponents while preserving the rest of the assembly. The authored rack uses nine approved teaching groups; generated hidden structures remain labeled as inferred unless supported by exact-model references.

- **Make explanations spatial

Team